Recall past decisions, sessions, and saved conversations from your dropdat capsule library and save new ones via any MCP-capable AI client.
dropdat_recall — hybrid vector + full-text semantic search over capsules, with optional tag and limit filters.
dropdat_read — fetch a capsule's full message contents by id, optionally including lineage.
dropdat_list — browse recent capsules, optionally filtered by tag.
dropdat_capsule — save the current conversation or a selected slice as a new capsule with title, messages, tags, summary, source, and source URL.
dropdat_autocapsule — import a full verbatim Claude Code .jsonl transcript as a capsule without passing messages.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@@dropdat/mcp recall capsules about natural language processing"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP server for dropdat. Lets any MCP-capable AI client (Claude Code, Cursor, Cline, Claude Desktop) recall, read, and save capsules in the user's dropdat library.
Related MCP server: Runar
The Python demo exposes the same five tools through FastMCP and routes execution with LangGraph. It supports a local SQLite demo and the existing Go API. See its README for setup and the stdio walkthrough.
Tool | Purpose |
| Hybrid vector + full-text search over the capsule library. |
| Fetch one capsule's full contents by id (optional lineage). |
| Browse recent capsules, optional tag filter. |
| Save the current conversation slice as a new capsule (model picks the messages). |
| Import filtered user/assistant text and nested tool-result text from a Claude Code |
dropdat_recall calls the hybrid /capsules/search endpoint
(vector cosine ranking + PostgreSQL full-text ranking fused via RRF). Set OPENAI_API_KEY on the API for
full semantic recall; without it the endpoint falls back to full-text search.
# from npm (recommended — clients can also `npx -y @dropdat/mcp`)
npm install -g @dropdat/mcp
# or from source
git clone https://github.com/dropdat/mcp.git
cd mcp && npm install && npm run buildIssue an API key in the dashboard → API Keys (token shown once,
shape dk_live_…). Export it:
export DROPDAT_API_KEY=dk_live_xxx
# Optional — defaults to https://dropdat.app, the hosted API.
# Set this only if you're running the API locally or self-hosted.
# export DROPDAT_API_BASE=http://localhost:8080If you're self-hosting the API and have DEV_AUTH_BYPASS=1 enabled,
the API ignores tokens entirely and assumes DEV_USER_ID. Use a real
key against any normal deployment.
Add to ~/.claude/mcp.json (or project .mcp.json):
{
"mcpServers": {
"dropdat": {
"command": "npx",
"args": ["-y", "@dropdat/mcp"],
"env": {
"DROPDAT_API_KEY": "dk_live_xxx"
}
}
}
}Same shape — npx -y @dropdat/mcp as the command, DROPDAT_API_KEY
in env. Transport is stdio.
npm run dev # tsx, no rebuild
npm run build # emit dist/The compiled dist/index.js starts with a #!/usr/bin/env node
shebang and is chmod +x'd on build — once published, clients can
npx -y @dropdat/mcp instead of pinning a path.
All against the Go API under /api/v1:
GET /capsules?q=&tag=&limit=
POST /capsules/search
GET /capsules/{id}
GET /capsules/{id}/lineage
POST /capsules
Bearer auth — accepts either a Clerk session JWT or a dk_* API key.
The MCP server uses the API-key path so it survives long-running
agent sessions without a refresh dance.
Save the FULL verbatim Claude Code session as a capsule by reading its on-disk .jsonl transcript directly. Use this instead of dropdat_capsule when the user wants the entire conversation, not just a model-reconstructed slice. The server reads the file itself — no need to pass messages.
Inputs: title (required), summary, tags, transcriptPath, +2 more.
Save the current conversation (or a relevant slice of it) as a new capsule. Call when the user says 'remember this', 'save this conversation', or at the end of a meaningful session.
Inputs: title (required), messages (required), summary, source, +2 more.
List recent capsules, optionally filtered by tag. Use to browse, not to search.
Inputs: tag, limit.
Fetch one capsule's full contents (all messages) by id. Pair with dropdat_recall to drill into a hit.
Inputs: id (required), includeLineage.
Hybrid (vector + keyword) semantic search across the user's dropdat capsule library. Use this whenever the user asks about prior decisions, past sessions, or anything they may have discussed with another AI — even if their phrasing doesn't match exact words from earlier conversations. Returns ranked hits with id, title, summary, tags, and a fused score.
Inputs: query (required), tag, limit.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 5 tools
Each tool has a clear and distinct purpose: autocapsule saves full sessions, capsule saves current slice, list browses, read fetches by ID, and recall searches. No overlap or ambiguity.
All tools use the 'dropdat_' prefix followed by a verb (autocapsule, capsule, list, read, recall). 'autocapsule' is a compound but still readable; overall pattern is consistent.
5 tools is well-scoped for a capsule management server. Each tool covers an essential operation without unnecessary redundancy.
Core CRUD-like operations are present: create (capsule, autocapsule), read (list, read), and search (recall). Missing delete or update may cause minor gaps, but the primary workflow is covered.
Search, read, cite, create, and safely update a user's private KeepFlash knowledge library.
Read-only memory search and task context for AI clients. Requires a Mem.Center Pro account.
Agentic search over your Dewey document collections from any MCP-compatible client.
Discover, compare, route, and execute machine-accessible capabilities for AI agents.
